Fix tokenizer extra_special_tokens crash and stabilize transformers

fix: Use correct unsloth gemma-4 model identifiers

feat: Add percentage progress to downloading status in UI and extension

feat: Add download status polling to UI and Firefox extension

feat: Switch models to use Gemma 4 from unsloth collection

feat: Add support for switching models and the 27b-4a backend

feat: Add Global Saliency Calculation Mode

fix: remove preprompt text from final API response and exclude preprompt from score normalization

Save study.db to writable /tmp or /data directory

Redirect root to index.html for HF Space

fix: ensure tokens without whitespace group correctly and update hardcoded html

feat: finalized FlowRead AI with hardcoded user study and polished UI

Highlight 40 percent of words for maximum saliency in study

Lower threshold for user study to highlight more salient words

Hardcode study HTML to reduce API load

Initial commit: text saliency pro with Gemma 2B